微博用户的兴趣及性格分析

发布时间：2019-01-13 08:54

【摘要】：随着互联网的快速普及,人们从一个信息匮乏的时代过渡到信息爆炸的时代。在这样一个时代中,如何更为准确地获取所需要的信息资讯将是一个难点。微博是一种新型的兼具社交网络服务和互联网媒体功能的平台。它可以帮助用户更加实时地获取或者发布信息。基于这种新型的媒体平台提供更为个性化的服务,将会进一步帮助用户更为准确和实时地获取感兴趣的内容。而在此之前,系统必须先分析用户兴趣爱好等信息。本文就是针对该研究热点,基于微博文本对用户信息进行分析。本文主要对新浪微博平台上的微博文本进行综合分析。其研究特色在于仅使用微博文本为分析的数据对象,从多个方面进行分析用户的信息。在本文中主要从微博文本中所蕴含用户的兴趣、情感和性格三个方面来进行分析,进而获取用户的信息。第一个方面是从微博文本中分析用户的兴趣,使用宏观和微观两个层面来表示用户的兴趣。通过三元过滤法来消除直接使用主题模型的训练结果中词项与主题之间关系矩阵中所可能含有的杂质,然后再使用WTMR(Word-themes mutual reinforcement)模型获取单条微博文本的主题概率分布。在获取单条微博文本主题概率分布的基础上,再一次使用WTMR模型获取微博文本集合的主题概率分布,即该微博用户兴趣信息的主题概率分布。同时,使用伪长文本和贪心策略,来提取用户感兴趣的关键词用来明确用户感兴趣的话题对象。整合用户兴趣的主题概率分布和关键词,从而获取用户兴趣。实验证明该方法能较为准确和简便地提取微博文本中所蕴含的用户兴趣。第二个方面主要是分析微博文本中的情感。改进主题模型中的经典方法LDA(Latent Dirichlet Allocation),提出DLDA(Double Latent Dirichlet Allocation)模型。将文本所蕴含的语义和情感视为平等关系来改进LDA。与此同时,使用不同的词权重来优化Gibbs Sampling。从实验结果可以看出,该方法可以更为准确地分析微博文本中所蕴含的情感极性。第三个方面是分析微博文本中所蕴含微博用户的性格。基于SC-LIWC词典,获取微博文本中所蕴含的用户性格。使用SC-LIWC词典和Big Five Model理论获取微博文本中每个词对应的大五类性格因素值。然后,再利用WTMR模型,从微博文本中挖掘用户的性格信息。从实验的结果可知,本文的方法可以较为准确地计算微博文本中蕴含用户性格。本文着重从多个方面分析微博文本中所蕴含的用户的兴趣、情感和性格内容。为了更好地展示本文研究的成果,本文设计一个基于上述三个研究结果的个性化微博服务系统。其包括三种不同类型的个性化服务,分别是:个性化微博排序、文字云导读和用户性格分析助手。这三种服务,除了给用户提供更高效地获取信息的功能,还可以用来分析用户的性格信息。
[Abstract]:With the rapid popularity of the Internet, people transition from an era of information scarcity to an era of information explosion. In such an era, how to obtain the information needed more accurately will be a difficult point. Weibo is a new type of social network services and Internet media features of the platform. It can help users to obtain or publish information in real time. Providing more personalized services based on this new media platform will further help users to obtain more accurate and real-time content of interest. Before this, the system must first analyze the user's interests and other information. This article is aimed at this research hot spot, based on Weibo text carries on the analysis to the user information. This article mainly carries on the comprehensive analysis to the Sina Weibo platform Weibo text. The characteristic of the research is that only Weibo text is used as the data object to analyze the user's information from many aspects. In this paper, we analyze the user's interest, emotion and character in Weibo's text, and then get the user's information. The first aspect is to analyze the user's interest from Weibo's text and express the user's interest at macro and micro level. In this paper, the possible impurities in the relation matrix between the words and the subject are eliminated in the training result of the direct use of the topic model by using the three element filter method, and then the topic probability distribution of the single Weibo text is obtained by using the WTMR (Word-themes mutual reinforcement) model). On the basis of obtaining the subject probability distribution of a single Weibo text, this paper uses WTMR model again to obtain the topic probability distribution of Weibo text set, that is, the topic probability distribution of the user's interest information. At the same time, pseudo-long text and greedy strategy are used to extract the keywords of interest to identify the subject of interest to the user. Integrating the topic probability distribution and key words of user interest to obtain user interest. Experiments show that this method can extract user interest in Weibo text accurately and easily. The second aspect is to analyze the emotion in Weibo's text. The DLDA (Double Latent Dirichlet Allocation) model is proposed by improving the classical method of topic model LDA (Latent Dirichlet Allocation),. Improve LDA. by treating the semantic and emotional implications of the text as equal relations At the same time, use different word weights to optimize Gibbs Sampling. The experimental results show that this method can more accurately analyze the emotional polarity contained in Weibo's text. The third aspect is to analyze the character of the user contained in Weibo's text. Based on the SC-LIWC dictionary, the user character contained in Weibo's text is obtained. The SC-LIWC dictionary and Big Five Model theory are used to obtain the five types of personality factors corresponding to each word in Weibo's text. Then, the WTMR model is used to mine the user's character information from Weibo's text. The experimental results show that the proposed method can accurately calculate the user character in Weibo's text. This paper analyzes the user's interest, emotion and character in Weibo's text. In order to better show the results of this study, this paper designs a personalized Weibo service system based on the above three research results. It includes three different types of personalized services: personalized Weibo sorting, text cloud reading and user character analysis assistant. In addition to providing users with more efficient access to information, these three services can also be used to analyze users' personality information.
【学位授予单位】：上海大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP393.092;TP391.1

【相似文献】